fix: make pyaudio an optional [client] extra so headless installs work

[MushiSenpai]

Move pyaudio to an optional client extra so headless / slim / API-only installs work.

pyaudio is a core dependency, but it needs the system PortAudio headers
(portaudio19-dev) to build — so pip install of fish-speech fails on slim
containers, CPU-only boxes, and any environment that only needs the API/server, not
local audio playback. PyAudio is used in exactly one place: native streaming playback
in tools/api_client.py.

Changes

• pyproject.toml — remove pyaudio from core dependencies; add a client
  optional-dependency group (client = ["pyaudio"]).
• tools/api_client.py — drop the module-level import pyaudio; import it lazily
  inside the --streaming playback branch, with a friendly ImportError pointing to
  pip install -e .[client].

No behavior change for users who already have PyAudio installed; headless/slim installs
now succeed, and only hit the requirement if they actually use streaming playback.

(Same spirit as the input-validation hardening in #1303.)

---

Disclosure: drafted with AI assistance (Claude); reviewed and submitted by me. The commit carries a Co-Authored-By trailer.

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[MushiSenpai]

Still applies to current main: pyaudio is still a core dependency (pyproject.toml:40), but it is only used for --play streaming playback in tools/api_client.py. That means every headless / slim / CPU / API-only install has to build PortAudio for a client-only feature, which is a common cause of install failures in containers.

This moves it to an optional [client] extra and makes the import lazy behind the playback branch, with a friendly ImportError pointing at pip install -e .[client] — so the playback path still works for anyone who wants it. Happy to rebase if useful.

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[MushiSenpai]

Re-verified against current main (214da3cd): this still applies.

pyaudio is still a core dependency at pyproject.toml:40, and it is only needed for --play streaming
playback in the client. That makes every headless, slim or API-only install build PortAudio, or fail.

The PR moves it into a client extra, using the [project.optional-dependencies] section that already
exists at :52, so pip install fish-speech stays lean and pip install "fish-speech[client]" keeps
playback. No conflicts, still mergeable. Happy to rename the extra if you prefer a different name.